Speaker
Description
This study investigates the application of Encoder–Processor–Decoder architectures, which have been successfully employed in large machine learning–based weather forecasting models, for wind power prediction. Using the Anemoi framework from the European Centre for Medium-Range Weather Forecasts (ECMWF), we evaluate graph-based neural networks over a domain covering Belgium and the Belgian Offshore Zone (BOZ) with a forecast horizon of up to 48 hours. These models are trained to forecast key meteorological variables, including wind speed, wind direction, and temperature, derived from the Copernicus Regional Reanalysis for Europe (CERRA). Wind power is treated as a diagnostic variable, and verification is performed using publicly available production data from the European Network of Transmission System Operators for Electricity (ENTSO-E). We assess how both input feature selection and network architecture affect prediction accuracy.
